Right before the news was dominated by fears of an AI apocalypse, the publishing industry was beset by multiple AI scandals. From the cancellation of a high-profile horror novel to the collapse of a multi-million dollar book deal, concerns of AI use reigns supreme in every case.
Ultimately, though, I don’t think AI is an insurmountable problem in the publishing industry (the very pressing concerns of rogue superintelligence, however, is way beyond my pay grade). Here are three things that need to happen to fix this problem.
Step One: Define Exactly What Constitutes AI Involvement
Here’s the good news: AI-generated content is a pretty well-understood red line in the publishing community. Royal Road defines AI-generated “…as where the author is prompting the AI to write the content and is merely directing it.” And here’s a bit more good news: I think most people - writers, publishers and readers - understand that passing off AI-generated text as one’s own work is a serious problem.
But after that, things get dicey. Royal Road goes to define AI assistance as “…using an AI tool as where it’s the author’s creativity and structure but it uses the AI’s voice and tone. To put it simply, where you take a paragraph that is already written and ask an AI to improve it for you.” (Royal Road also has a separate category for spell-checkers like Grammerly under General Assistive Technology, which generally aren’t frowned upon).
When I first started drafting Fang Trap, I did not understand this distinction. I would write a chapter and give it to ChatGPT for feedback and the a dopamine boost. ChatGPT would tell me what worked, but occasionally, it would find a sentence or a paragraph that was clunky or didn’t make sense. It would offer to clear it up, and I figured, “Sure, why not? It’s my words”.
What I didn’t understand was that I was directing the AI to manipulate the text, even on my behalf.
That was crossing into AI assistance.
Since I didn’t want an AI Assisted tag following me around on Royal Road, I quickly discontinued the practice. Fortunately, it was only the first few chapters of the initial draft of Fang Trap *-* and the story has been rewritten and revised several times over without any direct AI assistance.
Now, I use AI purely for brainstorming and alpha reading. Any changes AI recommends I make by hand, at my discretion.
I say this all, partially in the interest of transparency, but also to raise a particular point: I was fortunate to realize what constituted AI assistance relatively early on in the writing process. What happens when writers aren’t so lucky, and have been using AI assistance for drafts upon drafts - or even land agents or book deals without knowing the exact definition?
That’s why the definitions of AI-generated and AI-assisted content have to be explicitly defined and offered in tandem. It does authors no good to be confident they are not producing AI-generated content, only to face an AI-assisted pitfall later in their careers.
Step #2: Separate Good Faith Actors from Bad Faith Actors
AI is rapidly-evolving technology which has emerged in one of the most polarized moments in history. With that in mind, it shouldn’t be surprising that people can - and will - disagree on just how AI should be used.
Some people feel AI gives them tools and resources previously out of their reach.
Others have serious quandaries over the ethics of AI training, the speed of its development and the threat it poses to the environment.
There are real principles grounding each view, and these discussions should absolutely play out in the public sphere.
But for the purposes of publishing, transparency and honesty going forward matters just as much.
People will make different decisions when it comes to the use of AI. That shouldn’t be a mark of shame, nor should it spark a witch-hunt…so long as they do so openly and transparently.
Royal Road makes writers disclose AI-generated or AI-assisted content when they upload writing - as I previously mentioned. Amazon Kindle has a similar process - AI generation, defined as “as text, images, or translations created by an AI-based tool. If you used an AI-based tool to create the actual content (whether text, images, or translations), it is considered “AI-generated,” even if you applied substantial edits afterwards.” AI-generated content must be disclosed, while AI-assistant content does not.
These are good first steps, but to be effective, they need to be implemented industry-wide and treated as a definite standard within the publishing industry.
As it stands now, many publishers do have rules against submitting AI-generated content, and with good reason. But many fall short and don’t indicate what “using AI” constitutes. Is using AI for proofreading and brainstorming prohibited? What about AI tools like Grammarly?
Right now the industry needs a good boilerplate definition to guide prospective writers on the allowed parameters of AI use - and the consequences of withholding that information from agents, editors, publishers.
The goal should not be to eliminate the use of AI in publishing - it should be to establish the acceptable uses of AI. Reasonable people will continue to disagree whether AI should be allowed in writing at all, and that’s okay.
The goal shouldn’t be to settle that debate - it should be to separate those who are using AI openly and transparently within accepted parameters - and those who are lying to make a quick buck.
But how do we detect the latter category?
Don’t worry - I have some ideas on that too.
Step #3 - Make AI Detection a Career instead of a One-Stop App
Pangram is one of the most high-profile AI detection programs out there, and its results were often cited at the heart of each ongoing AI publishing scandal. But the app has also been questioned for the possibility of false positives, as well as the wisdom of using AI to detect AI.
I would also argue the stakes are just too high to get this wrong. In addition to the multi-million dollar deals terminated, incorrect AI assessments can also make it less likely publishers find new voices - or that new writers submit their work at all.
And to make matters worse, Stony Brook University estimates 20% of self-published books on Amazon are AI-generated. Something has to be done.
Long, long ago (I’m old now), I was a forum moderator on a website called FatWallet. Part of my job entailed stopping spammers from posting on the site. Some spammers were fairly obvious - first-time members posting very shady websites, with affiliate information baked into their URLs. Others required more research into their posting habits, the sites they were posting while reviewing readily available information.
And that’s my pitch for AI detection - make it a career instead of a one-and-done search app.
This isn’t to say services like Pangram won’t continue to be of use - they’ll often be the first line of defense against AI-generated content.
But putting trained specialists in the role of AI detection mods would help tamp down on abuse.
As it stands now, the risks of abuse within the publishing industry are almost just as damaging as unmitigated AI-generated writing. What if misinformation about AI use goes viral on an otherwise innocent author? What if accusers are ideologically motivated to push a narrative of AI-generated writing against an author? What happens then?
A team of human AI detection mods won’t solve all of these problems, but it puts guardrails in place while keeping humans - not apps - as the primary decision makers, while also creating an appeal system to give every accused author a fair shake at least.
AI detection moderators would likely work for publishers and platforms to ensure - to the best of their ability - that AI-generated content isn’t published in violation of preset AI transparency rules. Human AI mods probably won’t detect every piece of AI-generated content - nor should they be expected to - but they should make a good faith effort that makes AI-generated content less profitable, and over time, less prevalent.
Taken as a whole, these three steps won’t totally eliminate AI slop from making its way to readers, nor will it entirely shield the publishing industry from future AI scandals. What it will do, however, is eliminate any ambiguity going forward on the accepted (though probably still debated) use of AI and the consequences of not disclosing it.
On a whole, I think AI has the potential to genuinely assist both authors and publishers in beneficial ways, but for that to happen, we have to figure out where we want AI in our lives….and where we don’t. And while I don’t think there will be any consensus on that for quite some time, these steps can give us some ground rules going forward in a time when the line between AI and human authorship looks increasingly blurred.